- [Dependencies](#dependencies-23)
    - [Added](#added-23)
    - [Changed](#changed-23)
    - [Removed](#removed-23)
- [v1.19.0-beta.0](#v1190-beta0)
  - [Downloads for v1.19.0-beta.0](#downloads-for-v1190-beta0)
    - [Source Code](#source-code-23)
    - [Client binaries](#client-binaries-23)
    - [Server binaries](#server-binaries-23)
    - [Node binaries](#node-binaries-23)

pkg syscall (freebsd-arm64), const AF_INET6 = 28
pkg syscall (freebsd-arm64), const AF_INET6_SDP = 42
pkg syscall (freebsd-arm64), const AF_INET6_SDP ideal-int
pkg syscall (freebsd-arm64), const AF_IPX = 23
pkg syscall (freebsd-arm64), const AF_IPX ideal-int
pkg syscall (freebsd-arm64), const AF_ISDN = 26
pkg syscall (freebsd-arm64), const AF_ISDN ideal-int
pkg syscall (freebsd-arm64), const AF_ISO = 7

        including matmuls and convolutions, due to the use of
        [TensorFloat-32](https://blogs.nvidia.com/blog/2020/05/14/tensorfloat-32-precision-format/).
        Specifically, inputs to such ops are rounded from 23 bits of precision
        to 10 bits of precision. This is unlikely to cause issues in practice
        for deep learning models. In some cases, TensorFloat-32 is also used for